About Connection pooling
Learn how connection pooling works in Neon
Neon uses PgBouncer to support connection pooling, enabling up to 10,000 concurrent connections. PgBouncer is a lightweight connection pooler for Postgres.
How to use connection pooling
To use connection pooling with Neon, use a pooled connection string instead of a regular connection string. A pooled connection string adds the -pooler
option to your compute ID, as shown below:
The Connection Details widget on the Neon Dashboard provides Pooled connection checkbox that adds the -pooler
option to a connection string for you. You can copy a pooled connection string from the Dashboard or manually add the -pooler
option to the endpoint ID in an existing connection string.
info
The -pooler
option routes the connection to a connection pooling port at the Neon Proxy.
Connection limits without connection pooling
Each Postgres connection creates a new process in the operating system, which consumes resources. Postgres limits the number of open connections for this reason. The Postgres connection limit is defined by the Postgres max_connections
parameter. In Neon, max_connections
is set according to your compute size — and if you are using Neon's Autoscaling feature, it is set according to your minimum compute size.
Compute Size (CU) | vCPU | RAM | max_connections |
---|---|---|---|
0.25 | 0.25 | 1 GB | 112 |
0.50 | 0.50 | 2 GB | 225 |
1 | 1 | 4 GB | 450 |
2 | 2 | 8 GB | 901 |
3 | 3 | 12 GB | 1351 |
4 | 4 | 16 GB | 1802 |
5 | 5 | 20 GB | 2253 |
6 | 6 | 24 GB | 2703 |
7 | 7 | 28 GB | 3154 |
8 | 8 | 32 GB | 3604 |
9 | 9 | 36 GB | 4000 |
10 | 10 | 40 GB | 4000 |
The formula used to calculate max_connections
for Neon computes is RAM in bytes / 9531392 bytes
. For a Neon Free Plan compute, which has 1 GB of RAM, this works out to approximately 112 connections. Larger computes offered with paid plans have more RAM and therefore support a larger number of connections. For example, a compute with 12 GB of RAM supports up to 1351 connections. You can check the max_connections
limit for your compute by running the following query from the Neon SQL Editor or a client connected to Neon:
note
Seven connections are reserved for the Neon-managed Postgres superuser
account. For example, for a 0.25 compute size, 7/112 connections are reserved, so you would only have 105 available connections. If you are running queries from the Neon SQL Editor, that will also use a connection. To view connections that are currently open, you can run the following query:
Even with the largest compute size, the max_connections
limit may not be sufficient for some applications, such as those that use serverless functions. To increase the number of connections that Neon supports, you can use connection pooling. All Neon plans, including the Neon Free Plan, support connection pooling.
Connection pooling
Some applications open numerous connections, with most eventually becoming inactive. This behavior can often be attributed to database driver limitations, running many instances of an application, or applications with serverless functions. With regular Postgres, new connections are rejected when reaching the max_connections
limit. To overcome this limitation, Neon supports connection pooling using PgBouncer, which allows Neon to support up to 10,000 concurrent connections to the pooler endpoint.
The use of connection pooling, however, is not a magic bullet: As the name implies, connections to the pooler endpoint together share a pool of connections to the normal Postgres endpoint, so they still consume some connections to the main Postgres instance.
To ensure that direct access to Postgres is still possible for administrative tasks or similar, the pooler is configured to only open up to 64 connections to Postgres for each user to each database. For example, there can be only 64 active connections from role alex
to the neondb
database through the pooler. All other connections by alex
to the neondb
database will have to wait for one of those 64 active connections to complete their transactions before the next connection's work is started.
At the same time, role dana
will also be able to connect to the neondb
database through the pooler and have up to 64 concurrent active transactions across 64 connections, assuming the endpoint started with a high enough minimum Neon compute size to have a high enough max_connections
setting to support those 128 concurrent connections from the two roles.
Similarly, even if role alex
has 64 concurrently active transactions through the pooler to the neondb
database, that role can still start up to 64 concurrent transactions in the alex_db
database (a different database) when connected through the pooler; but again, only if the Postgres max_connections
limit can support the number of connections managed by the pooler.
For further information, see PgBouncer.
important
You will not be able to get interactive results from all 10,000 connections at the same time. Connections to the pooler endpoint still consume connections on the main Postgres endpoint: PgBouncer forwards operations from a role's connections through its own pool of connections to Postgres, and adaptively adds more connections to Postgres as needed by other concurrently active role connections. The 10,000 connection limit is therefore most useful for "serverless" applications and application-side connection pools that have many open connections but infrequent and short transactions.
PgBouncer
PgBouncer is an open-source connection pooler for Postgres. When an application needs to connect to a database, PgBouncer provides a connection from the pool. Connections in the pool are routed to a smaller number of actual Postgres connections. When a connection is no longer required, it is returned to the pool and is available to be used again. Maintaining a pool of available connections improves performance by reducing the number of connections that need to be created and torn down to service incoming requests. Connection pooling also helps avoid rejected connections. When all connections in the pool are being used, PgBouncer queues a new request until a connection from the pool becomes available.
Neon PgBouncer configuration settings
Neon's PgBouncer configuration is shown below. The settings are not user-configurable, but if you are a paid plan user and require a different setting, please contact Neon Support. For example, Neon sometimes raises the default_pool_size
setting for users who support a large number of concurrent connections and repeatedly hit PgBouncer's pool size limit.
The following list describes each setting. For a full explanation of each parameter, please refer to the official PgBouncer documentation.
pool_mode=transaction
: The pooling mode PgBouncer uses, set totransaction
pooling.max_client_conn=10000
: Maximum number of client connections allowed.default_pool_size=64
: Default number of server connections to allow per user/database pair.max_prepared_statements=0
: Maximum number of prepared statements a connection is allowed to have at the same time.0
means prepared statements are disabled.query_wait_timeout=120
: Maximum time queries are allowed to spend waiting for execution. Neon uses the default setting of120
seconds.
Connection pooling in transaction mode
As mentioned above, Neon uses PgBouncer in transaction mode (pool_mode=transaction
), which limits some functionality in Postgres. Functionality NOT supported in transaction mode includes:
SET
/RESET
LISTEN
WITH HOLD CURSOR
PREPARE / DEALLOCATE
PRESERVE
/DELETE ROWS
temp tablesLOAD
statement- Session-level advisory locks
These session-level features are not supported transaction mode because:
- In this mode, database connections are allocated from the pool on a per-transaction basis
- Session states are not persisted across transactions
Avoid using SET statements over a pooled connection
Due to the transaction mode limitation described above, users often encounter issues when running SET
statements over a pooled connection. For example, if you set the Postgres search_path
session variable using a SET search_path
statement over a pooled connection, the setting is only valid for the duration of the transaction. As a result, a session variable like search_path
will not remain set for subsequent transactions.
This particular search_path
issue often shows up as a relation does not exist
error. To avoid this error, you can:
- Use a direct connection string when you need to set the search path and have it persist across multiple transactions.
- Explicitly specify the schema in your queries so that you don’t need to set the search path.
- Use an
ALTER ROLE your_role_name SET search_path TO <schema1>, <schema2>, <schema3>;
command to set a persistent search path for the role executing queries. See the ALTER ROLE.
Similar issues can occur when attempting to use pg_dump
over a pooled connection. A pg_dump
operation typically executes several SET
statements during data ingestion, and these settings will not persist over a pool connection. For these reasons, we recommend using pg_dump
only over a direct connection.
For the official list of limitations, refer to the "SQL feature map for pooling modes" section in the pgbouncer.org Features documentation.
Connection pooling with schema migration tools
We recommend using a direct (non-pooled) connection string when performing migrations using Object Relational Mappers (ORMs) and similar schema migration tools. With the exception of recent versions of Prisma ORM, which support using a pooled connection string with Neon, using a pooled connection string for migrations is likely not supported or prone to errors. Before attempting to perform migrations over a pooled connection string, please refer to your tool's documentation to determine if pooled connections are supported.
Optimize queries with PgBouncer and prepared statements
Protocol-level prepared statements are supported with Neon and PgBouncer as of the PgBouncer 1.22.0 release. Using prepared statements can help boost query performance while providing an added layer of protection against potential SQL injection attacks.
Understanding prepared statements
A prepared statement in Postgres allows for the optimization of an SQL query by defining its structure once and executing it multiple times with varied parameters. Here's an SQL-level example to illustrate. Note that direct SQL-level PREPARE
and EXECUTE
are not supported with PgBouncer (see below), so you can't use this query from the SQL Editor. It is meant to give you a clear idea of how a prepared statement works. Refer to the protocol-level samples below to see how this SQL-level example translates to different protocol-level examples.
fetch_plan
here is the prepared statement's name, and $1
acts as a parameter placeholder.
The benefits of using prepared statements include:
- Performance: Parsing the SQL and creating the execution plan happens just once, speeding up subsequent executions. This performance benefit would be most noticeable on databases with heavy and repeated traffic.
- Security: By sending data values separately from the query, prepared statements reduce the risk of SQL injection attacks.
You can learn more about prepared statements in the PostgreSQL documentation. See PREPARE.
Use prepared statements with PgBouncer
Since pgBouncer supports protocol-level prepared statements only, you must rely on PostgreSQL client libraries instead (direct SQL-level PREPARE
and EXECUTE
are not supported). Fortunately, most PostgreSQL client libraries support prepared statements. Here are a couple of examples showing how to use prepared statements with Javascript and Python client libraries:
Need help?
Join our Discord Server to ask questions or see what others are doing with Neon. Users on paid plans can open a support ticket from the console. For more details, see Getting Support.